HR: 0800h
AN: H11B-0295 [Abstracts]
TI: Characterization of Unsaturated Soil Properties Using A Markov Chain Monte Carlo Approach
AU: Robinson, B
EM: robinson@lanl.gov
AF: Hydrology, Geochemistry, and Geology Group, Los Alamos National Laboratory, MS T003, Los ALamos, Nm
87545
United States
AU: * Lu, Z
EM: zhiming@lanl.gov
AF: Hydrology, Geochemistry, and Geology Group, Los Alamos National Laboratory, MS T003, Los ALamos, Nm
87545
United States
AU: Higdon, D
EM: dhigdon@lanl.gov
AF: Statistical Sciences Group, Los Alamos National Laboratory, MS F600, Los Alamos, NM 87545
United States
AU: Newman, B
EM: bnewman@lanl.gov
AF: Atmospheric Climate and Environmental Dynamics, Los Alamos National Laboratory, MS J495, Los Alamos, NM
87545
United States
AB:
We consider transient flow and solute transport in an unsaturated heterogeneous soil column under infiltration and assume
that the constitutive relationships between unsaturated hydraulic conductivity vs. pressure head and the water content vs.
pressure head follow the van Genuchten-Mualem model. The parameters that characterize the van Genuchten-Mualem model are
measured at some spatial locations. In addition to these direct measurements, time-dependent observations on state variables
(pressure head, moisture content, and solute concentration or tracer travel time) are also available at some locations. The
aim is to develop a general technique to estimate the infiltration rate and the spatial distributions of soil parameters, as
well as the uncertainties associated with these estimates, given an arbitrary sampling of different kinds of measured data D.
We do so using a Markov Chain Monte Carlo approach (MCMC). Each soil parameter is represented (parameterized) by the
combination of some basis kernels centered at fixed spatial locations. The prior distribution for the vector of coefficients
$\theta$ in this combination is specified and some hyperparameters $\lambda_\theta$ in this distribution are updated. The
samples for the vector $\theta$ are then taken from a posterior distribution $\pi(\theta, \lambda_\theta|D)$. Starting from
any initial setting, realizations of a Markov chain are generated by updating only one component of q at a time according to
Metropolis rules. The posterior mean of the parameter $\theta$ (and thus the infiltration and the soil properties conditional
to all observations) can be estimated from the Markov chain realizations (ignoring some early realizations). The
uncertainties associated with the mean quantities can also be assessed from these realizations. In addition, the MCMC
approach provides an alternative for estimating conditional predictions of state variables and their associated
uncertainties. Numerical tests for flow in a hypothetic random porous medium show that estimated soil properties from the
MCMC approach are close to the original hypothetical random fields. These tests also reveal the relative worth of various
types of data in constraining the estimates of infiltration rate and transport velocity. This information can be used to
optimize data collection activities in a field setting.
DE: 3210 Modeling
DE: 3260 Inverse theory
DE: 1869 Stochastic processes
DE: 1875 Unsaturated zone
SC: Hydrology [H]
MN: 2004 AGU Fall Meeting